New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
TensorFlow library was compiled to use AVX2 instructions, but these aren't available on your machine. #1181
Comments
Thanks, however there are several things I need to comment. The documentation of https://github.com/ashwan1/django-deepspeech-server is wrong, you don't need the Please file issue only for actual reproductible and verified issed on DeepSpeech. if you are just reaching for help, please use Discourse: https://discourse.mozilla.org/c/deep-speech |
@lissyx Actually I mean to run requirements.txt of my project. But, thanks for pointing out, if it was not clear for you, it might not be for others. I fixed that. :) |
@ashwan1 - Your documentation wasn't wrong at all. I found it easy to understand. Thanks for changing it though. :)
|
@lissyx - An error message with "aborted" as the severity is not a harmless message. The content of the message doesn't say that "the CPU is capable of AVX2". It states that - "The TensorFlow library was compiled to use AVX2 instructions, but these aren't available on your machine.", which means that the CPU is NOT capable of AVX2. That was also confirmed in the cpuinfo, where the flag for AVX was present, but the flag for AVX2 was not present.
Have now run deepspeech this morning and the message appears again ..
I tried to determine the version of TensorFlow
Yet Tensorflow is in the Deepspeech requirements.txt file, and the installation of Deepspeech had no errrors, as follows:
Thanks for the link. I did try this wheel the other day - https://index.taskcluster.net/v1/task/project.deepspeech.deepspeech.native_client.master.cpu/artifacts/public/deepspeech-0.1.0-cp35-cp35m-manylinux1_x86_64.whl I'm using pip3 and the version is 9.0.1, and from the changelog, that appears to be the latest. |
Have been doing a bit of research on this problem. Some related links, with possibly the last one the most relevant for Ubuntu users: https://discourse.mozilla.org/t/error-after-following-installation-steps/22578 https://github.com/lakshayg/tensorflow-build yaroslavvb/tensorflow-community-wheels#26 |
@jehoshua7 I misread your AVX2 error message: this is an error that has been explained multiple times. You need to use newer artifacts from TaskCluster. On my system, ubuntu 17.10, |
@lissyx - okay, that is great you have the same system. I waited for 17.10.1 as there was a CPU problem (but I don't use a Lenovo). I checked pip3 and python3 within the virtual environment and outside of it, just to be sure. Both of them are: pip 9.0.1 from /usr/lib/python3/dist-packages (python 3.6) I then download this artifact - https://index.taskcluster.net/v1/task/project.deepspeech.deepspeech.native_client.master.cpu/artifacts/public/deepspeech-0.1.0-cp36-cp36m-manylinux1_x86_64.whl , and then installed it
Requirement already satisfied: deepspeech==0.1.0 from file:///home/********/Downloads/deepspeech-0.1.0-cp36-cp36m-manylinux1_x86_64.whl in ./lib/python3.6/site-packages and then tried DeepSpeech again
2018-01-19 21:04:00.937657: F tensorflow/core/platform/cpu_feature_guard.cc:35] The TensorFlow library was compiled to use AVX2 instructions, but these aren't available on your machine. |
@jehoshua7 Can you try that with a clean virtualenv? We have not yet bumped version number. And you have not forced And please, until proven otherwise, this is unlikely to be a DeepSpeech code issue, so this discussion should move to Discourse, were there is already a lot of documentation for the very same issue: https://discourse.mozilla.org/c/deep-speech |
Yes, I was just thinking a clean installation, as the DeepSpeech /requirements.txt has
yet it seems tensorflow is not installed ? As you can see by a post above, the installation went okay. So, I will remove everything and start again.
Mozilla Discourse - Have lost count of the number of times I have attempted to 'register' there. It's not the normal, name, username/email and pwd. Everytime I try the link they send in an email, I get some message about authentication failing. I could contact someone at Discourse, ..but hmm, I'd need to login to do that as there is absolutely no 'contact' info, except some sort of critical/emergency scenario. So I added a post to Mozilla forums, a mod there said they are not 'connected', different mods,admin. Eventually I looked through the mods at Discourse and one did have an email address. Have emailed that person, so will wait and see. Just wanting to let you know I'm doing a lot to try and get registered there. (Yes they do have a github type register/sign up, but it assumes one has a two part security setup via a mobile phone. Not everyone has a mobile phone, ..lol ) |
@jehoshua7 I do insist, there is no need for TensorFlow Python package for inference, the |
@lissyx - Okay thanks. I assumed that error message was sourced from tensorflow, as the msg content had tensor flow in it.
No doubt a lot of people would have attempted to 'sign up' at Discourse, and then given up. I don't give up easily. :)
It may be caused by a number of factors. All our email goes through a server which has "MailScanner" and in the emails , all had the message link - "MailScanner has detected a possible fraud attempt from "auth.mozilla.auth0.com" claiming to be Sign in to discourse.mozilla.org". Does MailScanner (and similar tools) change the authentication link ? Of course we should be having this conversation elsewhere, a forum where we can have a discourse. But that is catch-22, is it not ? (tongue in cheek). Another problem in the email has "The link will expire in five minutes.". Our email server is USA and I assume the one at mozilla.org is also, but some of the emails arrived too late. It needs to be 30 mins minimum. This 'about' is well hidden - https://discourse.mozilla.org/about , but the one moderator who has an email address public has replied to my email and copied in an admin from https://discourse.mozilla.org/about , so that is a good start. Where to from here ?? Of interest is that I do have an 'account' at https://support.mozilla.org , as that has a normal registration process. Not sure where to discuss this ? I did have a mod reply to a post at https://support.mozilla.org , so that is good. I have now actually been able to login to Discourse, BUT everytime I need to login, it requires the "passwordless" authentication procedure. It's still 'hit and miss' as to whether there will be another successful login. |
Have just done the clean installation, after removing the virtual environment. Same error message. Forced an upgrade, tried deepspeech again, same error message. :( |
@jehoshua7 Well, I don't know what is wrong. https://index.taskcluster.net/v1/task/project.deepspeech.deepspeech.native_client.master.cpu/artifacts/public/deepspeech-0.1.0-cp36-cp36m-manylinux1_x86_64.whl works for sure on AVX-only CPUs like yours. Besides, does the error message still mentions lack of Also, it would help a lot if you gave clear and detailed documentation of each of your steps when you say |
The debug commands didn't display anything extra ? Found that library, it is within the virtual environment where deepspeech is installed. Here are the terminal commands and output ..
Used Dolphin and deleted /home/********/Public/Servers and all sub paths
|
If I was running Android it would let me off with a "warning" ,..lol // If the CPU feature isn't present, log a fatal error. |
Several hours of going through posts on many sites, and found this one at #1023 (comment)
So tried that ..
What, no message ? So my question is, was there anything that was installed/changed today that contributed to it working, or was it simply a case of obtaining/installing that wheel ? I did 'part' perform a build today, but only got so far with it. |
@jehoshua7 It was The fact that you had Anyway, it is all working, as I said since the begining, so nothing more to add here. |
@lissyx Thanks for the tip regards using
Not possible at all. This is a 'bare bones' computer and I have only worked within the virtual environment. If I had installed deepspeech outside of any virtual environment, then deepspeech would show. Out of those 3 python packages, the virtual environment shows: deepspeech (0.1.0) and outside the virtual environment, pip3 shows: numpy (1.12.1) and that is showing in the base system because python3-numpy is installed by default. Related post at https://stackoverflow.com/questions/42339034/python-module-in-dist-packages-vs-site-packages
Looking back through the logs I have kept for the deepspeech install, that version of scipy came from the wheel that was recommended
So at least we know the source of that version of scipy
Yes, and thanks for all your help, much appreciated. I ran a small test WAV on deepspeech and the word error rate is 43.63 % , but that's another question. :) |
No, the 0.19.1 scipy version confirms that the wrong, old-AVX2 DeepSpeech was still installed. |
@lissyx - thanks. Some things will remain a mystery. :) |
@jehoshua7 I install the CPU version from your last link, |
This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs. |
I had installed DeepSpeech and also a DeepSpeech server ( https://github.com/ashwan1/django-deepspeech-server ). Went to start the server and got an error message - "2018-01-17 08:21:49.120154: F tensorflow/core/platform/cpu_feature_guard.cc:35] The TensorFlow library was compiled to use AVX2 instructions, but these aren't available on your machine.
Aborted (core dumped)"
Apparently I need to compile TensorFlow on the same computer. Is there a list somewhere to match Kubuntu 17.10.1 and a HP Probook 4330S please ?
I did try the wheel file at https://getpocket.com/redirect?url=https%3A%2F%2Findex.taskcluster.net%2Fv1%2Ftask%2Fproject.deepspeech.deepspeech.native_client.master.cpu%2Fartifacts%2Fpublic%2Fdeepspeech-0.1.0-cp35-cp35m-manylinux1_x86_64.whl&formCheck=c926fe8ba25e2f144816b163a4aab101 , but when I went to install it, got an error message - "msg - "deepspeech-0.1.0-cp35-cp35m-manylinux1_x86_64.whl is not a supported wheel on this platform."
Here is some information on the computer that I'm using:
=========================
HP Probook 4330S
Software
Kubuntu 17.10.1
Kernel: 4.13.0-25-generic
OS Type: 64 bit
Hardware
Processors: 4 x Intel Core i5-2430M CPU @ 2.40GHz
Memory: 3.8 GiB of RAM
Each processor has max speed: 3000
Supported instruction sets:
Intel MMX
Intel SSE
Intel SSE2
Intel SSE3
Intel SSE4
and the (cutdown) cpuinfo is:
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe syscall nx rdtscp lm constant_tsc arch_perfmon pebs bts rep_good nopl xtopology nonstop_tsc cpuid aperfmperf pni pclmulqdq dtes64 monitor ds_cpl vmx est tm2 ssse3 cx16 xtpr pdcm pcid sse4_1 sse4_2 x2apic popcnt tsc_deadline_timer aes xsave avx lahf_lm epb pti tpr_shadow vnmi flexpriority ept vpid xsaveopt dtherm ida arat pln pts
AVX but no AVX2
$ gcc --version
gcc (Ubuntu 7.2.0-8ubuntu3) 7.2.0
I'm new to Python, so please excuse my ignorance. When I do a "pip3 list' I can see DeepSpeech but no TensorFlow ?? Yet if I look in the requirements.txt for DeepSpeeech, TensorFlow is there.
Have done all this in a virtual environment, so dropped out of there (deactivate) and di another "pip3 list", but still no Tensorflow showing.
The text was updated successfully, but these errors were encountered: